“…Mehmood et al [ 45 ] proposed to use the Viola-Jones algorithm to crop the detection features and use lightweight deep learning to extract features for an efficient and fast retrieval matching process. To ensure the security of image retrieval, Li et al [ 46 ] combined a deep convolutional neural network and hashing algorithm for extracting image features and used vector homomorphic encryption to improve search effectiveness. Sui et al [ 47 ] designed a conditional adversarial network with which the noise affection according to the reconstructed image can be reduced greatly.…”